Tenstorrent’s $1.5 Billion Round Moves the Test to Deployment
Fresh equity backs manufacturing and commercial expansion, while earlier financing and software disclosures clarify what still needs proving.
Share
Posts by tag
Inference
24 posts
Modal Labs Is Reportedly Near a $750 Million Round, but the Deal Is Not Closed
A TechCrunch source says Modal Labs is nearing a $750 million round at a $15.75 billion post-investment valuation; the company has not confirmed the terms.
Share
OpenAI Claims Jalapeño Chip Raises Inference Efficiency
OpenAI's first Jalapeño benchmarks claim higher user throughput and efficiency, but meaningful deployment is not expected until 2027.
Share
Etched’s $21B Valuation Turns Inference Hardware Into a High-Stakes Bet
Etched has raised $700 million at a $21 billion valuation as it expands from accelerator chips into complete systems for frontier-model inference.
Share
Groq’s $350M Pivot Turns a Chip Challenger Into a Nvidia-Powered Neocloud
Groq’s $350 million round funds a rapid neocloud expansion after the former Nvidia challenger abandoned a pure chip strategy, but capacity growth brings a tougher economics test.
Share
Kog Wants Software—not New Chips—to Unlock Faster AI Inference
French startup Kog claims deep GPU optimisation can accelerate single-request AI inference, but its most dramatic benchmark still relies on a purpose-built small model.
Share
OpenAI’s Ultrafast Preview Pushes Frontier Models Toward Real-Time Work
OpenAI says a Cerebras-powered preview can run GPT-5.6 Sol up to 14 times faster, though access is limited and practical performance remains unproven.
Share
Chinese AI Models Turn Cost Into a Competitive Weapon
Alibaba, DeepSeek and Moonshot show why sparse architecture, task-level consumption and open weights matter more than headline token prices alone.
Share
SenseTime’s Galaxy Project Tackles China’s Multi-Chip AI Problem
Galaxy links domestic accelerators, software and data centres into one ecosystem, but its throughput and energy claims still need independent testing.
Share
Kimi K3 Moves China’s AI Bottleneck From Compute to Memory
Moonshot’s 2.8-trillion-parameter open-weight model activates only a fraction at once, but running it still demands enormous memory and infrastructure.
Share